What is feature reuse in DenseNet and how does it differ from ResNet?

DenseNet connects each layer to all previous layers via concatenation, enabling explicit feature reuse, unlike ResNet which uses addition.

DenseNet introduces dense connectivity where each layer receives inputs from all previous layers. This encourages feature reuse and reduces vanishing gradients. Unlike ResNet’s additive skip connections, DenseNet concatenates feature maps, preserving all intermediate representations. This leads to efficient parameter usage and strong gradient flow.